Automated Tool for Question Generation

1. A computer-implemented method for generating questions from a source document, comprising: selecting passages of text from the source document based on a first criterion; transforming the selected text passages based on coreference analysis; selecting fragments of text in the transformed text passages based on matching combined semantic-syntactic patterns from a pattern library; and automatically generating the questions by transforming the selected text fragments.

2. The method of claim 1 , wherein the first criterion is that, to be selected, a given passage of text has a similarity greater than or equal to a content relevance threshold, relative to at least one subject matter descriptor of the source document.

3. The method of claim 1 , wherein the transformation of a given selected text passage based on coreference analysis further comprises: identifying, in the given text passage, coreferences comprising one or more of: anaphora, cataphora, endophora, or exaphora; classifying the coreferences according to a plurality of coreference types; and for an instant type which is a most easily resolved one of the coreference types: selecting the coreferences classified as the instant type; analyzing at least the given text passage to determine most probable resolutions of the selected coreferences; and replacing one or more of the selected coreferences with their most probable resolutions.

4. The method of claim 3 , further comprising, subsequent to the replacing: regenerating the given text passage; selecting new coreferences classified as a second one of the coreference types which is a most easily resolved type after the instant type; analyzing at least the given text passage to determine most probable resolutions of the new selected coreferences; and replacing the new selected coreferences with their most probable resolution.

5. The method of claim 3 , wherein a sequence of the selecting coreferences, the analyzing, and the replacing is repeated for a plurality of successive instant types of the coreference types.

6. The method of claim 1 , wherein at least one of the combined semantic-syntactic patterns in the pattern library comprises a plurality of nodes representing respective syntactic parts of a text fragment, at least one of the nodes having a corresponding semantic attribute, and at least one pair of the nodes connected by a relationship attribute.

7. The method of claim 6 , wherein a given one of the combined semantic-syntactic patterns has a degree in the range 10 - 20 inclusive, wherein the degree comprises a sum of numbers of: nodes, semantic attributes, syntactic attributes, and relationship attributes, over all the nodes of the given combined semantic-syntactic pattern.

8. The method of claim 6 , wherein the matching comprises: comparing the nodes, the corresponding semantic attribute(s), and the relationship attribute(s) of a given one of the combined semantic-syntactic patterns with a given text fragment to determine a matching score; and selecting the given text fragment as matching the given combined semantic-syntactic pattern if the matching score is greater than or equal to a matching threshold.

9. The method of claim 1 , wherein the transformation of a given selected text fragment is dependent on a first pattern of the pattern library matched by the given selected text fragment, and results in a first one of the questions.

10. The method of claim 9 , further comprising: selecting a selector based on the first pattern; and applying the selector to the given selected text fragment to determine a correct answer for the first question.

11. The method of claim 10 , further comprising generating internal distractors for the first question by performing one or more of: applying negation to the correct answer; applying shifting to the correct answer; or applying a pattern-substitution on the correct answer.

12. The method of claim 10 , further comprising: identifying text, in the source document or in an external corpus, having a syntactic match with the correct answer and a non-synonymous semantic similarity between vectors representing the identified text and the correct answer; and transforming the identified text into a distractor.

13. One or more computer-readable media storing instructions which cause the method of claim 1 to be performed, when the instructions are executed by one or more computer processors.

14. A method of generating questions from a source document, comprising: at one or more computers: receiving the source document; progressively refining text of the source document in phases to generate the questions, the phases comprising: a first selection phase, a first transformation phase at which semantic content of individual sentences is localized, a second selection phase based on matching combined semantic-syntactic patterns from a pattern library, and a second transformation phase at which the questions are generated; and outputting the generated questions; wherein one or more of the combined semantic-syntactic patterns in the pattern library comprises a plurality of nodes representing respective syntactic parts of a text fragment, at least one of the nodes having a corresponding semantic attribute, and at least one pair of the nodes connected by a relationship attribute; and wherein a given one of the one or more combined semantic-syntactic patterns has a degree of at least 10, wherein the degree comprises a sum of numbers of: nodes, semantic attributes, syntactic attributes, and relationship attributes, over all the nodes of the given combined semantic-syntactic pattern.

15. The method of claim 14 , wherein the second selection phase comprises reverse parsing by matching an input to the second selection phase against the pattern library of combined semantic-syntactic patterns.

16. The method of claim 14 , wherein at least one of the combined semantic-syntactic patterns spans more than one sentence.

17. The method of claim 14 , wherein the first selection phase comprises determining respective content values for portions of the source document, and selecting, based on the content values, a first one of the portions as output of the first selection phase.

18. The method of claim 17 , wherein the first selection phase, the first transformation phase, the second selection phase, and the second transformation phase are performed in that order.

19. The method of claim 18 , further comprising one or more pairs of an additional selection phase and an additional transformation phase.

20. A computer-implemented method for generating one or more questions, comprising: receiving one or more source documents; analyzing the source document(s); wherein the analyzing comprises determining respective first numerical content value scores for portions of the source document(s); first selecting one or more of the portions based upon the respective first numerical content value scores; resolving coreferences in the first selected portion(s); first transforming the first selected portion(s) into one or more second version documents comprising transformations of the resolved coreferences; reverse parsing the second version document(s) using a pattern library to second select one or more fragments of the second version document(s); second transforming the second selected fragment(s) into the question(s).